X-Git-Url: http://git.veekun.com/zzz-pokedex.git/blobdiff_plain/d131ce6a550d4d3572584c04984550d151eccd7b..095fdd7d58c012cbe6345d369d00c26b8159e935:/pokedex/lookup.py diff --git a/pokedex/lookup.py b/pokedex/lookup.py index b8dc1a1..768a674 100644 --- a/pokedex/lookup.py +++ b/pokedex/lookup.py @@ -14,7 +14,7 @@ from whoosh.qparser import QueryParser import whoosh.scoring import whoosh.spelling -from pokedex.util import namedtuple +from pokedex.compatibility import namedtuple from pokedex.db import connect import pokedex.db.tables as tables @@ -69,10 +69,11 @@ class LanguageWeighting(whoosh.scoring.Weighting): # Apply extra weight weight = weight * self.extra_weights.get(text, 1.0) - if doc['language'] == None: + language = doc.get('language') + if language is None: # English (well, "default"); leave it at 1 return weight - elif doc['language'] == u'Roomaji': + elif language == u'Roomaji': # Give Roomaji a little boost; it's most likely to be searched return weight * 0.9 else: @@ -81,8 +82,9 @@ class LanguageWeighting(whoosh.scoring.Weighting): class PokedexLookup(object): - INTERMEDIATE_LOOKUP_RESULTS = 25 - MAX_LOOKUP_RESULTS = 10 + MAX_FUZZY_RESULTS = 10 + MAX_EXACT_RESULTS = 43 + INTERMEDIATE_FACTOR = 2 # The speller only checks how much the input matches a word; there can be # all manner of extra unmatched junk, and it won't affect the weighting. @@ -102,6 +104,7 @@ class PokedexLookup(object): tables.Move, tables.Nature, tables.Pokemon, + tables.PokemonForm, tables.Type, ) ) @@ -174,7 +177,13 @@ class PokedexLookup(object): display_name=whoosh.fields.STORED, # non-lowercased name ) - if not os.path.exists(self.directory): + if os.path.exists(self.directory): + # create_in() isn't totally reliable, so just nuke whatever's there + # manually. Try to be careful about this... + for f in os.listdir(self.directory): + if re.match('^_?(MAIN|SPELL)_', f): + os.remove(os.path.join(self.directory, f)) + else: os.mkdir(self.directory) self.index = whoosh.index.create_in(self.directory, schema=schema, @@ -204,34 +213,33 @@ class PokedexLookup(object): # Add the basic English name to the index if cls == tables.Pokemon: - # Pokémon need their form name added - # XXX kinda kludgy - add(row.full_name, None, u'en', u'us') - - # If this is a default form, ALSO add the unadorned name, - # so 'Deoxys' alone will still do the right thing - if row.forme_name and not row.forme_base_pokemon_id: - add(row.name, None, u'en', u'us') - else: - add(row.name, None, u'en', u'us') + # Don't re-add alternate forms of the same Pokémon; they'll + # be added as Pokémon forms instead + if not row.is_base_form: + continue + elif cls == tables.PokemonForm: + if row.name: + add(row.pokemon_name, None, u'en', u'us') + continue # Some things also have other languages' names # XXX other language form names..? - for foreign_name in getattr(row, 'foreign_names', []): - moonspeak = foreign_name.name - if row.name == moonspeak: - # Don't add the English name again as a different + seen = set() + for language, name in getattr(row, 'name_map', {}).items(): + if name in seen: + # Don't add the name again as a different # language; no point and it makes spell results # confusing continue + seen.add(name) - add(moonspeak, foreign_name.language.name, - foreign_name.language.iso639, - foreign_name.language.iso3166) + add(name, language.name, + language.iso639, + language.iso3166) # Add Roomaji too - if foreign_name.language.name == 'Japanese': - roomaji = romanize(foreign_name.name) + if language.identifier == 'ja': + roomaji = romanize(name) add(roomaji, u'Roomaji', u'ja', u'jp') writer.commit() @@ -288,47 +296,53 @@ class PokedexLookup(object): name = name.strip() prefixes = prefix_chunk.split(',') - user_valid_types = [_.strip() for _ in prefixes] + user_valid_types = [] + for prefix in prefixes: + prefix = prefix.strip() + if prefix: + user_valid_types.append(prefix) # Merge the valid types together. Only types that appear in BOTH lists # may be used. # As a special case, if the user asked for types that are explicitly - # forbidden, completely ignore what the user requested - combined_valid_types = [] - if user_valid_types and valid_types: - combined_valid_types = list( - set(user_valid_types) & set(combined_valid_types) - ) - - if not combined_valid_types: - # No overlap! Just use the enforced ones - combined_valid_types = valid_types - else: - # One list or the other was blank, so just use the one that isn't - combined_valid_types = valid_types + user_valid_types + # forbidden, completely ignore what the user requested. + # And, just to complicate matters: "type" and language need to be + # considered separately. + def merge_requirements(func): + user = filter(func, user_valid_types) + system = filter(func, valid_types) + + if user and system: + merged = list(set(user) & set(system)) + if merged: + return merged + else: + # No overlap; use the system restrictions + return system + else: + # One or the other is blank; use the one that's not + return user or system - if not combined_valid_types: - # No restrictions - return name, [], None + # @foo means language must be foo; otherwise it's a table name + lang_requirements = merge_requirements(lambda req: req[0] == u'@') + type_requirements = merge_requirements(lambda req: req[0] != u'@') + all_requirements = lang_requirements + type_requirements # Construct the term - type_terms = [] lang_terms = [] - final_valid_types = [] - for valid_type in combined_valid_types: - if valid_type.startswith(u'@'): - # @foo means: language must be foo. - # Allow for either country or language codes - lang_code = valid_type[1:] - lang_terms.append(whoosh.query.Term(u'iso639', lang_code)) - lang_terms.append(whoosh.query.Term(u'iso3166', lang_code)) - else: - # otherwise, this is a type/table name - table_name = self._parse_table_name(valid_type) + for lang in lang_requirements: + # Allow for either country or language codes + lang_code = lang[1:] + lang_terms.append(whoosh.query.Term(u'iso639', lang_code)) + lang_terms.append(whoosh.query.Term(u'iso3166', lang_code)) - # Quietly ignore bogus valid_types; more likely to DTRT - if table_name: - type_terms.append(whoosh.query.Term(u'table', table_name)) + type_terms = [] + for type in type_requirements: + table_name = self._parse_table_name(type) + + # Quietly ignore bogus valid_types; more likely to DTRT + if table_name: + type_terms.append(whoosh.query.Term(u'table', table_name)) # Combine both kinds of restriction all_terms = [] @@ -337,7 +351,7 @@ class PokedexLookup(object): if lang_terms: all_terms.append(whoosh.query.Or(lang_terms)) - return name, combined_valid_types, whoosh.query.And(all_terms) + return name, all_requirements, whoosh.query.And(all_terms) def _parse_table_name(self, name): @@ -379,7 +393,7 @@ class PokedexLookup(object): results.append(LookupResult(object=obj, indexed_name=record['name'], name=record['display_name'], - language=record['language'], + language=record.get('language'), iso639=record['iso639'], iso3166=record['iso3166'], exact=exact)) @@ -470,12 +484,26 @@ class PokedexLookup(object): ### Actual searching - searcher = self.index.searcher() - # XXX is this kosher? docs say search() takes a weighting arg, but it - # certainly does not - searcher.weighting = LanguageWeighting() - results = searcher.search(query, - limit=self.INTERMEDIATE_LOOKUP_RESULTS) + # Limits; result limits are constants, and intermediate results (before + # duplicate items are stripped out) are capped at the result limit + # times another constant. + # Fuzzy are capped at 10, beyond which something is probably very + # wrong. Exact matches -- that is, wildcards and ids -- are far less + # constrained. + # Also, exact matches are sorted by name, since weight doesn't matter. + sort_by = dict() + if exact_only: + max_results = self.MAX_EXACT_RESULTS + sort_by['sortedby'] = (u'table', u'name') + else: + max_results = self.MAX_FUZZY_RESULTS + + searcher = self.index.searcher(weighting=LanguageWeighting()) + results = searcher.search( + query, + limit=int(max_results * self.INTERMEDIATE_FACTOR), + **sort_by + ) # Look for some fuzzy matches if necessary if not exact_only and not results: @@ -510,12 +538,8 @@ class PokedexLookup(object): ### Convert results to db objects objects = self._whoosh_records_to_results(results, exact=exact) - # Only return up to 10 matches; beyond that, something is wrong. We - # strip out duplicate entries above, so it's remotely possible that we - # should have more than 10 here and lost a few. The speller returns 25 - # to give us some padding, and should avoid that problem. Not a big - # deal if we lose the 25th-most-likely match anyway. - return objects[:self.MAX_LOOKUP_RESULTS] + # Truncate and return + return objects[:max_results] def random_lookup(self, valid_types=[]): @@ -526,8 +550,10 @@ class PokedexLookup(object): table_names = [] for valid_type in valid_types: table_name = self._parse_table_name(valid_type) - # Skip anything not recognized. Could be, say, a language code - if table_name: + # Skip anything not recognized. Could be, say, a language code. + # XXX The vast majority of Pokémon forms are unnamed and unindexed, + # which can produce blank results. So skip them too for now. + if table_name and table_name != 'pokemon_forms': table_names.append(table_name) if not table_names: @@ -535,26 +561,20 @@ class PokedexLookup(object): # were valid, but this function is guaranteed to return # *something*, so it politely selects from the entire index instead table_names = self.indexed_tables.keys() - - # Rather than create an array of many hundred items and pick randomly - # from it, just pick a number up to the total number of potential - # items, then pick randomly from that, and partition the whole range - # into chunks. This also avoids the slight problem that the index - # contains more rows (for languages) for some items than others. - # XXX ought to cache this (in the index?) if possible - total = 0 - partitions = [] - for table_name in table_names: - count = self.session.query(self.indexed_tables[table_name]).count() - total += count - partitions.append((table_name, count)) - - n = random.randint(1, total) - while n > partitions[0][1]: - n -= partitions[0][1] - partitions.pop(0) - - return self.lookup(unicode(n), valid_types=[ partitions[0][0] ]) + table_names.remove('pokemon_forms') + + # Pick a random table, then pick a random item from it. Small tables + # like Type will have an unnatural bias. The alternative is that a + # simple search for "random" will do some eight queries, counting the + # rows in every single indexed table, and that's awful. + # XXX Can we improve on this, reasonably? + table_name = random.choice(table_names) + count = self.session.query(self.indexed_tables[table_name]).count() + id, = self.session.query(self.indexed_tables[table_name].id) \ + .offset(random.randint(0, count - 1)) \ + .first() + + return self.lookup(unicode(id), valid_types=[table_name]) def prefix_lookup(self, prefix, valid_types=[]): """Returns terms starting with the given exact prefix.